datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
aegis-anti-sycophancy
🛡️ AEGIS Anti-Sycophancy Preference Dataset (Expanded 2,562 Pairs)
The AEGIS Anti-Sycophancy Preference Dataset is an alignment dataset curated to eliminate sycophancy, stance reversals, and ungrounded flattery in LLMs and LLM-as-a-Judge systems under user pressure.
📊 Dataset Overview
Total Pairs: 2,562 preference pairs (2,305 train / 257 test)
Features:
user_input: High-pressure, authoritative, empathetic, or logical user prompts pushing false claims.
chosen:… See the full description on the dataset page: https://huggingface.co/datasets/Hooshaai/aegis-anti-sycophancy.aegis-gsm8k-bench
🧮 AEGIS GSM8K Adversarial Robustness Benchmark
Preprocessed GSM8K benchmark splits formatted in native Apache Parquet, configured for multi-turn adversarial stress testing and counterfactual invariance evaluation in LLM-as-a-Judge research.
📊 Dataset Structure & Splits
Train Split: 7,473 mathematical reasoning problems with explicit reference derivations.
Test Split: 1,319 problems used for out-of-distribution adversarial debate evaluation.
Format:… See the full description on the dataset page: https://huggingface.co/datasets/Hooshaai/aegis-gsm8k-bench.aegis-bilingual-industrial-ai-dataset
AEGIS AI Bilingual Industrial Operations Dataset
AEGIS AI Bilingual Industrial Operations Dataset is a synthetic English–Arabic dataset designed for experimentation with multilingual enterprise AI systems, Retrieval-Augmented Generation (RAG), industrial question answering, document intelligence, semantic search, and AI workflow automation.
The dataset extends the original AEGIS AI industrial dataset with structured Arabic and English representations while preserving industrial… See the full description on the dataset page: https://huggingface.co/datasets/syed7741/aegis-bilingual-industrial-ai-dataset.aegis-multilingual-guard-dataset-v0.1-samples
AEGIS Multilingual Guard Dataset — Public Sample (100 per domain)
📋 This is a small public preview of the full AEGIS Multilingual Guard Dataset, a multilingual red-team / guardrail corpus. It contains 100 representative records per business domain (12 domains → 1,200 records) so you can explore the schema, languages and label balance before requesting the full dataset.
⚠️ Safety-research data. Records contain adversarial attack prompts (jailbreaks, prompt injections… See the full description on the dataset page: https://huggingface.co/datasets/YATAV-ENT/aegis-multilingual-guard-dataset-v0.1-samples.aegis-training-v2.3
AEGIS RevOps Training Dataset v2.3
World-class training data for the AEGIS 3-tier Revenue Operations AI assistant.
v2.3 Improvements (World-Class Release)
New Capabilities
Computed-Truth Business Math: 600+ examples with verifiable numeric answers (pipeline rollups, win rates, commissions, churn cohorts)
Adversarial Reasoning: 350+ examples handling contradictory signals, causal traps, and incomplete data
Noisy Real-World Inputs: 250+ examples with missing… See the full description on the dataset page: https://huggingface.co/datasets/atr0p05/aegis-training-v2.3.aegis-stressbench-trajectories
🛡️ AEGIS-StressBench Trajectories & Behavioral Automata Archive
This repository stores empirical multi-turn debate trajectories, behavioral transition matrices, and evaluation logs for AEGIS-StressBench.
📁 Repository Contents
sample_trajectories.jsonl: Complete multi-turn conversational traces recording step-by-step judge rationales, adversary attacks across 4 personas, and outcome classifications.
fsm_transitions.json: Empirical 9-state Finite… See the full description on the dataset page: https://huggingface.co/datasets/Hooshaai/aegis-stressbench-trajectories.aegis-training-v2.1
AEGIS RevOps Training Dataset v2.1
High-quality, domain-focused training data for the AEGIS 3-tier Revenue Operations AI assistant.
v2.1 Improvements (over v2)
✅ 91% RevOps-relevant (Main model) - up from 35%
✅ Removed domain dilution - filtered generic math/reasoning
✅ Added router hard negatives - 138 calibration examples
✅ Added RevOps anchors - 40+ true-to-AEGIS examples
✅ Clean Voice model - No <think> blocks in voice data
Dataset Description… See the full description on the dataset page: https://huggingface.co/datasets/atr0p05/aegis-training-v2.1.Aegis-Safety-DPO
Aegis: PolarAI's safety alignment dataset
Overview
Aegis-Safety-DPO is a high-density, (mostly) manually-curated preference dataset designed for Direct Preference Optimization (DPO) and Group Relative Policy Optimization (GRPO).
Unlike traditional safety datasets that train models to be "preachy," "evasive," or "apologetic", Aegis trains models that refuse to answer
Analyze the malicious request deeply using Chain-of-Thought (<think>).
Conclude objectively why the… See the full description on the dataset page: https://huggingface.co/datasets/PolarAI/Aegis-Safety-DPO.NYXIS-AEGIS-Knowledge
NYXIS-AEGIS Knowledge Base 🛡️
A curated collection of 500+ articles covering coding, mathematics, science, history, geography, and common knowledge — built to power the NYXIS-PRO reasoning engine.
What's Inside
Category
Topics
Examples
Coding
50+
Python, JavaScript, Data Structures, Algorithms, Docker, Git
Math
27
Calculus, Linear Algebra, Statistics, Probability, Logic
Science
25+
Quantum Mechanics, Genetics, CRISPR, Black Holes
History
18+
World Wars… See the full description on the dataset page: https://huggingface.co/datasets/QuantaSparkLabs/NYXIS-AEGIS-Knowledge.aegis-training-v2
AEGIS RevOps Training Dataset v2
High-quality training data for the AEGIS 3-tier Revenue Operations AI assistant.
Dataset Description
This dataset contains training examples for three specialized models:
Router (0.5B Model)
Purpose: Intent classification for routing queries
Intents: voice_simple, crm_lookup, complex_analysis, action_confirm, fallback
Examples: ~2,600
Voice (7B Model)
Purpose: Quick, conversational responses
Focus: Concise RevOps… See the full description on the dataset page: https://huggingface.co/datasets/atr0p05/aegis-training-v2.Aegis-Safety-DPO
Aegis: PolarAI's safety alignment dataset
Overview
Aegis-Safety-DPO is a high-density, (mostly) manually-curated preference dataset designed for Direct Preference Optimization (DPO) and Group Relative Policy Optimization (GRPO).
Unlike traditional safety datasets that train models to be "preachy," "evasive," or "apologetic", Aegis trains models that refuse to answer
Analyze the malicious request deeply using Chain-of-Thought (<think>).
Conclude objectively why the… See the full description on the dataset page: https://huggingface.co/datasets/2etatg/Aegis-Safety-DPO.Aegis-SFT-v3-Dataset
Aegis SFT v3 Training Dataset
7,361 (AI text → Student text) pairs for training an AI text humanizer.
Dataset Description
Each sample is a chat-formatted conversation where the model learns to rewrite formal AI-generated text into natural student voice.
Construction
Source: PERSUADE 2.0 Corpus — 25,996 pre-ChatGPT student essays
Filter: Holistic essay score 4-6, word count 100-600 → 7,361 essays
AI Rewrite: Each student essay was rewritten in… See the full description on the dataset page: https://huggingface.co/datasets/changcheng967/Aegis-SFT-v3-Dataset.aegis-training-v2.2
AEGIS RevOps Training Dataset v2.2
High-quality, domain-focused training data for the AEGIS 3-tier Revenue Operations AI assistant.
v2.2 Improvements (over v2.1)
Main model expanded: 2,025 examples (up from 825)
+1,200 synthetic RevOps examples across 8 categories:
Pipeline coverage (200)
Win rate analysis (150)
Commission calculations (150)
Churn analysis (150)
Forecasting (150)
Quota planning (100)
Territory management (100)
Table-based prompts (200)
Voice bridging… See the full description on the dataset page: https://huggingface.co/datasets/atr0p05/aegis-training-v2.2.smolified-aegis-inspector
🤏 smolified-aegis-inspector
Intelligence, Distilled.
This is a synthetic training corpus generated by the Smolify Foundry.
It was used to train the corresponding model ankan288/smolified-aegis-inspector.
📦 Asset Details
Origin: Smolify Foundry (Job ID: 34bfbeb2)
Records: 8962
Type: Synthetic Instruction Tuning Data
⚖️ License & Ownership
This dataset is a sovereign asset owned by ankan288.
Generated via Smolify.ai.